Content-based image retrieval (CBIR) has become an important research topic in the last two decades. A lot of visual feature based descriptors have been evolved, and several image matching algorithms have been developed based on such feature representations. In this article, we are proposing an image retrieval approach intended to search unknown landmarks using images captured or stored in handheld devices such as mobile phones or tabs. The proposed image retrieval approach is built on Position Scale Orientation-Scale Invariant Feature Transform (PSO-SIFT) based feature descriptor analysis. Principal component analysis has been used here for dimensionality reduction by filtering prominent features, and the image similarity search is computed over this feature space. The model also uses multiple appropriate images similar to the query image from the device database by extracting visual information based contextual saliency. Identifying appropriate images for retrieval can help to find suitable image features and could reduce the bandwidth requirement by limiting the number of transmitted features. This technique also helped to increase the retrieval performance. Image retrieval performance of the proposed method is analyzed using different standard benchmarks and are compared with state-of-the-art image retrieval techniques, and the results show that the algorithm performs well in terms of precision and overall recall rate.
Volume 12 | Issue 3